1. Introduction

View the most recent version.

Knowledge spillovers associated with the diffusion of new technologies have long been viewed as important drivers of modern economic growth (Rosenberg 1982, Landes 1998 and Romer 1990). Since knowledge spillovers both facilitate and are facilitated by regional economic agglomerations (Marshall 1920, Krugman 1991a and 1991b and Porter 1998) it is natural to consider whether knowledge spillovers also lead to agglomeration effects in the adoption of new technologies. Specifically, is the adoption of new technologies more highly agglomerated than other forms of economic activity and, if so, can this be explained by localized knowledge spillovers? However, the effect of knowledge spillovers on technology adoption has not been studied in the literature. The aims of this paper are to examine whether there are localized knowledge spillovers in the adoption of new technologies and to identify and estimate the effects of knowledge spillovers.

Figure 1 provides a simple answer to the first part of this question on the agglomeration of technology adoption. It shows that geographic concentration is higher for plants using advanced manufacturing technologies than it is for all plants in an industry.¹ This fact is new to the literature and raises an obvious, but important question: What explains the higher degree of geographic agglomeration among adopters of advanced-manufacturing technologies? One potential explanation points to knowledge spillovers across technology adopters. Unfortunately, this is not an easy hypothesis to explore. First, we rarely observe knowledge spillovers directly. Second, there are alternative potential explanations such as labour pooling, forward and backward linkages between suppliers and purchasers (Krugman 1991a; Fujita, Krugman and Venables 1999 and Porter 1990) and local amenities such as transportation infrastructure and weather, which have nothing to do with spillovers.²

In this paper, each of these explanations is investigated. While we find some support for almost all of them, our main result is that a plant is more likely to adopt a specific technology (e.g., a flexible manufacturing cell) if that specific technology has already been adopted by other plants in 'similar industries' in the same region. Similar industries here represent a self-constructed set of industries that shares a similar pattern of input purchases.³ This result cannot be traced back to a spurious correlation operating at the industry, region or industry × region levels. First, the result holds good, even after controlling for labour pooling, backward linkages to suppliers, forward linkages with buyers, and also industry-, region-, technology- and time-fixed effects. Second, the result shows that the effect is strongest with geographic, technological and functional similarities and it decays with the distance in those three dimensions. Third, the result is strongest when prior adopters are in a different industry than that of the potential adopter. In short, our findings strongly suggest the existence of communication across plants within the same geographical region.

Such communication implies that there are localized, learning-based knowledge spillovers (Case 1992; Jaffe, Trajtenberg and Henderson 1993; Powell and Brantley 1992 and von Hippel 1988). For instance, in the decision to adopt a new technology, potential adopters often face uncertainties about implementation costs. Since certain types of knowledge about the implementation of a new technology are tacit, learning about tacit knowledge is more likely to happen through direct observation of early adopters, demonstration, word-of-mouth and other informal mechanisms. Hence, the local presence of prior adopters would facilitate the rapid and complete diffusion of a new technology.

The analysis in this paper is based upon a proprietary panel data set on the adoption of 22 advanced-manufacturing technologies by 1,902 Canadian plants. I use these data to address the following questions. First, and most importantly, are there regional knowledge spillovers linking prior adopters to potential adopters? If so, does the extent of spillovers depend on the similarity between prior adopters and potential adopters where 'similarity' is measured in terms of the pattern of input purchases? Second, are knowledge spillovers from prior adopters to potential adopters conditional on geographic proximity? Are the effects of knowledge spillovers from prior adopters confined to potential adopters within the same geographical region or are they extended to geographically distant ones as well? Third, are knowledge spillovers bound within technological proximity? If one plant adopts technology τ , does this have any impact on other plants' adoption decision of any technology or only of technology τ ? Fourth, what is the sectoral scope of agglomeration externalities on technology adoption? That is, is it regional specialization in just a few industries (Marshall 1920) or regional diversification of industries (Jacobs 1970) that facilitates technology adoption?

While the effects of agglomeration on technology adoption are conjectured to be important in most discussions about agglomeration, there is very little related work. There are three relevant strands ofliterature. The first follows Jaffe, Trajtenberg and Henderson (1993), who studied the type of knowledge spillover that is captured by patent citations. The use of patent citations to study knowledge spillovers warrants further research on knowledge spillovers because of three particularities of patents. First, firms often do not patent (Levin et al. 1987 and Rosenberg 1982). Second, not all patents contain valuable information, hence they may not be the best measure of knowledge spillovers. Third, patents only describe a particular aspect of innovative knowledge, and not all innovative activities lend themselves to patenting. On the other hand, the 22 advanced-manufacturing technologies employed in this study do not share the above characteristics and are often general-purpose technologies that are of value and universally accessible. Consequently, they capture different aspects of knowledge than those obtained from patents, and hence they lend themselves as likely candidates for a study of knowledge spillovers and complement the literature of knowledge spillovers in a critical aspect.

The second strand in the literature is not about knowledge spillovers per se, but about the importance of different sources of agglomeration (Rosenthal and Strange 2001; Dumais, Ellison and Glaeser 1997 and Holmes 2002). These papers examined each source of agglomeration separately, but to the extent of considering knowledge spillovers, they either treated them as residuals or measured them imperfectly.

The third related strand of literature examines the impact of agglomeration on technology adoption. This literature consists of only two studies (Harrison, Kelley and Gant 1996; Kelley and Helper 1996) that examine the effects of location attributes on the adoption of computer numerically controlled (CNC) machines. As in the first and second strands, knowledge spillovers were not directly investigated nor were they isolated from the effects of other location attributes, since the goal was to examine the location attributes that better facilitated technology adoption. In addition, the literature warrants further research in this area because these are case studies in nature that are based on the adoption of one specific technology in a small number of plants in a small subset of industries—i.e., 342 plants in 21 3-digit Standard Industrial Classification (SIC) industries. Consequently, identification of the effects of knowledge spillovers is still left unanswered in the literature. This paper is aimed at overcoming these deficiencies by empirically identifying and separately estimating the impact of knowledge spillovers and other sources of agglomeration on technology adoption.

The paper's main finding is that technology adoption is facilitated by the presence of prior technology adopters with four characteristics: (1) they are adopters of the same technology, as opposed to adopters of advanced technology more generally; (2) they reside in the same geographical region; (3) they are similar to the potential adopter in that they purchase a similar set of intermediate goods and services; and (4) the effects of prior adopters are greatest if prior adopters are dissimilar to the potential adopter in that they do not operate in the same product market (i.e., the same 4-digit SIC code). This result holds good, even after controlling for the effects of regional labour pooling, regional linkages to suppliers and buyers, as well as industry-, region-, technology- and time-fixed effects. These findings are strongly indicative of the presence of localized-knowledge spillovers in the adoption of new technologies.

The remainder of this paper is organized as follows. Chapter 2 documents the higher concentration of technology-adopting plants than for all manufacturing plants. Chapter 3 discusses the methodology. Chapter 4 describes the data sources. Chapter 5 presents the results of agglomeration effects on technology adoption. Chapter 6 concludes.

1. We expand on this in Section 2.

2. See Hanson (2000) for discussion of the issues associated with identifying agglomeration effects.

3. Further details of the construction of similar industries are discussed in Section 4.2.